Whether it be the skeleton of a child found in the woods by police, the commingled skeletal remains of a family lost in a natural disaster, or bones recovered from a mass grave by human rights investigators, the forensic problem remains the same: How to identify the age, sex, and “population affinity” of bones from an infant, child, or adolescent?
Answering that question, according to specialists in forensic science and anthropology, is difficult because identifications of what are known as “subadults” are based on traditional forensic methods that lack the underlying data needed to accurately read markers of age, sex, and population affinity in skeletal remains. Accurate identification is important because more than 460,000 children are reported missing in the United States each year. Although the overwhelming majority are found or return home on their own, about 1 in 10,000 is not found alive, according to organizations focused on missing children.
Investigators confronted with subadult skeletal remains, defined as ranging from birth to approximately age 20, face several limits on what can be determined compared to examinations of adult skeletons. A key question when the remains of a young person are found is the age-at-death. The standard measures for determining the age-at-death of subadult skeletal remains include long bone lengths, teeth, and the fusion of growth plates in bones (epiphyseal fusion). But those indicators are only useful during specific periods of a person’s growth and, in isolation, are limited in what they can reveal.
“Compared with what we can do [in terms of the biological profile] with adults, there’s almost nothing we can do with subadults,” said Kyra Stull, an anthropologist at the University of Nevada, Reno. Stull has focused on developing a data-driven approach to identifying the age and sex of subadult skeletal remains and an algorithm specifically for age estimation in several research projects supported by the National Institute of Justice dating back to 2015. “With adults we have the capacity to estimate the population affinity [race, geographical origin], sex, stature, and age,” she said. “All of that information can really help us narrow down a potential list of missing individuals so law enforcement and identification units can go forward in trying to find a match.”
The problem of estimating age is much more challenging with subadults. The key issue is the error rate associated with forensic age estimations. While a five-year age range is good for adult skeletal remains — determining that a person was between ages 35 and 40 at death, for example — for subadults that five-year range equals a significantly larger relative error rate and loses much of its usefulness, Stull said.
One of the limiting factors in subadult age estimation is that most of the methods being used are based on longitudinal studies that ran from the 1920s to the 1950s. The reliance on old longitudinal studies is because there have not been enough samples of subadults to create a new approach, Stull said. “We’re basing age estimations on antiquated samples that are not reflective of the diverse America that we live in now,” she said. The purpose of those samples was for more general studies that tried to determine what was normal, or ideal, growth in children. The studies were intended to measure how some groups of children deviated from the ideal.
Stull’s response to the lack of accurate data for identifying the sex, age, and population affinity of subadults has been to develop the Subadult Virtual Anthropology Database (SVAD)[1] to significantly improve the examination of subadult skeletal remains. In a recent paper in the journal Forensic Sciences, Stull detailed the database, describing it as the “largest available repository of contemporary (2010-2019) subadult reference data from around the world.”[2]
The open access database, which contains data collected from 4,891 virtual images of subadults from eight countries, allows researchers to use an algorithm to analyze in detail the biological profiles of a wide variety of samples from around the world. SVAD is composed of CT scans and radiographs, many of them medical images, from 2,077 females and 2,814 males between birth and age 22. The images were gathered by Stull and her colleagues from Angola, Brazil, Colombia, France, the Netherlands, South Africa, Taiwan, and the United States.
Some are medical records from living subadults, and others are scans of deceased individuals. The data focus on five areas: length and breadth measurements of the six long bones in humans; vertebral/spinal canal measurements; pelvic measurements; dental development; and fusion of growth plates in bones. Data that are currently being collected, and will eventually also be freely available, include cranial measurements (craniometrics); cranial shape and form measurements, particularly nasal bone structure (macromorphoscopics); tooth size measurements; and the shape and form of teeth (dental morphology). Within each of these areas are extensive datasets based on multiple measurements of biological indicators, allowing an investigator with a set of skeletal remains to consider a wide range of markers for age, sex, and population affinity. The importance of any one of those markers shifts with the age-at-death of the remains, which makes the algorithm valuable for the entire subadult range.
Stull created the database because she realized that as a person moves through the stages of subadult life — newborn, infant, child, and adolescent — the relative importance of the different markers of age and sex change. “There is no best age indicator,” Stull said in a presentation about SVAD. “From birth to [age] 20 there are a lot of changes and depending on where you are [on the timeline] one may be more informative than another.”
Teeth may the best indicators of age when the mouth is full of teeth, she said, but “when you’re a 2-year-old you don’t have many teeth.” Long bones such as the femur, tibia, and humerus are good age markers up to ages 7-10, and epiphyseal fusion, the solidifying of growth plates in the bone, typically occurs in the teen years. The expression of sex in pelvic bones occurs around age 13 in females, which makes it possible to estimate the sex of individuals prior to their reaching maturity.
Funded by two grants from the National Institute of Justice and one from the National Science Foundation, SVAD is intended to encourage “forensic anthropological research specifically and forensic science research more generally towards open science.” Toward that goal, Stull has also developed KidStats and KSCollect, graphical user interfaces for researchers that are supported by the data in SVAD.[3] KidStats, developed with National Institute of Justice support, allows researchers to estimate the age of a subadult from specific changes on bones and by examining dental development using the new algorithm, the Mixed Cumulative Probit. KSCollect is a stand-alone open data entry program for subadult biological information that will be linked to KidStats soon. The end goal of all of this work, Stull said, “is to facilitate access to large-scale anthropological data … and limit the effects of missing data [and] structures that are common with physical skeletal collections.”
Much of Stull’s work in developing SVAD traces back to a project by Stephen Ousley, an anthropologist at Mercyhurst University in Erie, Pennsylvania. In a 2008 National Institute of Justice grant[4], Ousley, with Stull working as one of several co-investigators, created a radiographic database for modern American subadults. Based on 44,220 radiographic images of 9,709 subadults, the database was described as having “tremendous potential for research in age, sex, and ancestry estimation methods.”
Ousley’s study was an important first step in modernizing subadult research and moving away from the growth data studies of the 1920s as a basis for examining subadult skeletons. In his final report to NIJ,[5] Ousley noted the creation of a subadult database would “offer the most reliable sex and age standards with statistically determined age prediction intervals in the forensic setting,” and would spur development of new methods of analysis for forensic and clinical investigators.
Yet another researcher, Nicolas Herrmann, a forensic and dental anthropologist at Texas State University in San Marcos, Texas, is currently conducting an NIJ-supported project using transition analysis and machine learning methods to determine the age-at-death of subadults through dental root and crown development.[6] Herrmann noted that current methodology, which is based on tooth mineralization, often underestimates age by a year or more in subadults.
His project will focus on developing a standard set of teeth “to be scored for the most accurate age estimation, ancestry specific models, and potentially sex estimation from the timing of tooth development,” Herrmann wrote in a project summary. Machine learning will be used to further refine age estimates from dental development. The work resulted in a web-based analysis platform that can generate statistical age estimates for use in casework .
The overarching goal of the subadult research supported by the National Institute of Justice over the past decade has been to use evolving analytical approaches to create a database that allows investigators to accurately determine the age, sex, and population affinity of skeletal remains of infants, children, and adolescents. The work of Stull and other researchers has established, primarily through virtual imaging, skeletal reference collections of subadults that should significantly improve anthropological and forensic science research.